This chapter explores the concept of experienced optimization in hyper-parameter optimization, a critical task in Automatic Machine Learning (AutoML). Hyper-parameter optimization often involves derivative-free optimization (DFO) methods, which can be inefficient due to the high cost of evaluating hyper-parameter configurations. The chapter introduces an experienced optimization approach that leverages historical optimization data to improve efficiency in new tasks. Two algorithms, ExpSRacos and AdaSRacos, are presented, which utilize directional models trained on past optimization experiences to guide the search process. AdaSRacos further enhances this by adaptively selecting relevant historical experiences, ensuring that only useful information is utilized. The chapter includes empirical studies on synthetic and real-world hyper-parameter optimization tasks, demonstrating the effectiveness of the proposed methods in reducing evaluation costs and improving optimization performance. The results highlight the importance of experience adaptation in achieving efficient and effective hyper-parameter tuning.